200 likes | 305 Views
Mechanism Design for Keyword Auction. Wenjin Rong For CUHK, 2014. 09. 05. Two Questions. Who like advertising?. What kind of advertising do you like?. Some are Beautiful. Some other are annoying. What topic is today’s talk?. How to create “ beautiful ” ads?. √.
E N D
Mechanism Design for Keyword Auction Wenjin Rong For CUHK, 2014. 09. 05
Two Questions Who like advertising? What kind of advertising do you like?
What topic is today’s talk? How to create “beautiful” ads? √ Beautiful _ Good Looks: Branding Ads Beautiful _ Real Needs: Targeted Ads
Computational Advertising • What is Computational Advertising • Find the "best match" between a given user in a given context and a suitable advertisement . —— Broder and Dr. Vanja,2011 • What is Baidu? • Baidu is a high-tech company with mission to provide the best wayfor people to find information. Ads Best Match ∈ Baidu Mission
Advertising is A Kind of Matching In Which Channel Says What Who To Whom Feedback With What Effects Lasswell, 1948, The Structure And Function Of Communication In Society
Perfect Matching in Bipartite Graphs Ads slots Slot 1 Slot 2 Slot 3 Slot 4 Slot 5 Tutte, 1947, A Ring In Graph Theory; Hall, 1935, On Representatives Of Subsets
Efficient Matching Advertisers' Value Matrix • Efficient Matching:Maximum sum of each advertisers' Value 12+6+5=23 • But this result is unstable if there is no any constraint for advertisers. Tao Slot 1 Dong Slot 2 Hao Slot 3
Market Clearing Price Value Matrix、Profit Matrix and Price Price not to Clearing Market Demange et al, 1986, Multi-item Auction.
Advertisement Scheduling System:广告管家 date Ads Slots
VCGRealizes Market Values In Tao case: 1) When Tao is absent, all the other advertisers’utility is 4.6×0.5+1.8×0.2+1×0.1=2.76 2) When Tao is present, all the other advertisers’utilityis 4.6×0.2+1.8×0.1+1×0=1.1 3) The difference of both 1) and 2) is 2.76-1.1=1.66 4) So Tao must pay 1.66/0.5=3.32 for each click-through. Vickrey, 1961, Counterspeculation,Auctions and Competitive Sealed Tenders Clarke, 1971, Multipart Pricing of Public Goods Groves, 1973, Incentives in Teams
Generalized English Auction 0 0.5 6 1 5.5 1.5 5 2 4.5 4 2.5 3.5 3 Bergemann and Morris, 2004, Robust Mechanism Design
Deferred Acceptance M3>M1>M2 M1>M2>M2 M1>M2>M3 W2>W1>W3 M3>M1>M2 M3>M1>M2 M1>M2>M3 W2>W1>W3 W1 W2 W3 W1>W2>W3 W1>W2>W3 W1>W2>W3 M2 M1 M3 W1>W2>W3 W1>W2>W3 Shapley and Shubik,1972,The Assignment Game I: The Core
Generalized Second Price Auction(GSP) 5 Rank=1 CPC=4 • b=(3, 2, 1) is a Nash equilibrium. • But B can envy A: • if B replace A in slot 1, his payoff is (4-2)×0.1=0.2 > (4-1)×0.05=0.15 • Effective way to let off stream is raising bids. For example ,B raises his bid from $2 to $2.5: • if B replace A, his payoff is (4-2.5)×0.1=0.15 • so B should not want to “exchange” with the A , We call such vectors of bids “Locally Envy-Free.”. 4 Rank=2 CPC=2 2 Rank=Nothing CPC=0 Edelman et al,2005,Internet advertising and the generalized second price auction: Selling billions of dollars worth of keywords
Weighted GSP • Separation of CTR:CTRi j= qi ×ej position effect quality effect • Weighted GSP • Bid: Each advertiser bids an amount ba • Rank: Advertisers are ordered by qaba • b1 q1> b2 q2>…> bmqm • Price:psqs=bs+1 qs+1, Solving for pswe have Varian,2007,Position auctions
CTR Prediction • Logistic Regression Model • Problems: • Deep Learning Hinton & Salakhutdinov , 2006,Reducing the dimensionality of data with neural networks Bengio & LeCun, 2007, Scaling learning algorithms towards AI
Unified Auction Unified Auction wGSP Auction Modeling User Experience • max sum(fn(xn)) • s.t. sum(xn) <= ue_thr Abrams & Schwarz, 2008, Ad Auction Design and User Experience